Google launches Gemini 4 Argon API pricing

What did Google launch on October 13?

Google launched Gemini 4 Argon, its latest model, and made it globally available to developers and enterprises through the Gemini API, Google AI Studio and the Gemini platform. The model identifier is gemini-4-argon, and Google says it is the company’s latest and most capable model.

The rollout also extends beyond the developer API. Google said Google AI Ultra subscribers can use Argon in Google Antigravity, while Google AI Ultra and Pro subscribers can access it in the Gemini app and AI Mode in Search. Google Cloud said the model is available in Gemini for business on the web, desktop and mobile starting Oct. 13.

How does Gemini 4 Argon work for builders?

Google is positioning Argon as a long-context model built for multi-step work. Its documentation says the model uses a unified 1,048,576-token window, meaning input, thinking and output all share the same context budget. Google also says developers can tune thinking levels to low, medium or high, with medium as the default.

For long generations, Google says developers can use background execution with max_output_tokens above 65,536 to generate a single response past the usual per-request cap, up to 1 million tokens. The docs add that non-streaming REST generateContent calls have a 32,768-token per-request cap, while streaming calls are capped at 65,536 tokens. Google’s latest GenAI SDKs will automatically continue long decode requests when max_output_tokens exceeds the per-request limit.

Google also says explicit context caching is not supported on Argon, but implicit caching is enabled by default for prompts of at least 4,096 tokens.

What is Google charging for the API?

Google is offering Argon at an introductory rate of $2 per 1 million input tokens and $10 per 1 million output tokens through Nov. 30, 2026. The company said cached input tokens get a 95% discount, and its developer pricing page lists the introductory cached-input price at $0.10 per 1 million tokens.

After Nov. 30, Google said standard pricing will rise to $4 per 1 million input tokens and $20 per 1 million output tokens, with cached input tokens priced at $0.20 per 1 million tokens from Dec. 1, 2026. Google’s pricing page also lists discounted batch and flex rates of $1 per 1 million input tokens and $5 per 1 million output tokens through Nov. 30, 2026, plus priority pricing of $3.60 per 1 million input tokens and $18 per 1 million output tokens.

The model is not available on the Gemini API Free tier or the Google AI Pro plan.

Why is Google emphasizing Argon’s capabilities?

Google is pairing the launch with performance claims across software, science and safety workflows. In its announcement, the company said internal engineers had been shipping higher-quality code faster over the prior few weeks using Argon, and that use in Google Antigravity increased agentic code submissions by more than 35% while reducing code rollbacks by nearly a third.

Google also said the Chrome team used Argon on Chromium’s 40 million-line codebase to finish an architectural refactor across hundreds of files in weeks instead of months. Other reported examples include a 40% cut in in-memory session prep time and a 21% improvement in cold imports in google/adk-python, and a 2.85x inference speedup on Gemma 4 E2B via vLLM.

Outside software, Google said researchers used Argon to make the xTB open-source quantum chemistry package up to 2.3x faster at predicting molecular structures and 1.8x faster at conformer search. The company also said Isomorphic Labs is using Argon to accelerate drug discovery and design workflows by synthesizing complex PDFs and patents, reducing work from weeks to hours.

Google described additional demonstrations in biology, robotics and math, including five novel serine esterase designs for 4MU-acetate, an internal cornhole simulation that was transferred to Apptronik’s Apollo humanoid, and a Lean-verified arcsin implementation that Google says cut spacetime cost by more than 100x versus a published quantum CORDIC baseline. The company also said Argon scored 87% on its Embodied Reasoning benchmark and led the Gray Swan indirect prompt injection benchmark, cutting attack success across 15 adaptive attempts to 0.7%.

Google said the model’s external bio red teaming unlocked less than 1% of the biological subtasks the company tests for, and shortcut attempts on coding benchmarks dropped to 2%, its best result among previous Gemini models.